AI — Page 24

CoreWeave co-founder Brannin McBee wants investment-grade ratings to bring insurers into more chip-backed loans, with lower-rated AI-lab contracts the next test.

LightSeek says TokenSpeed produced the per-user result on four NVIDIA GB300 GPUs using NVFP4, but it has not disclosed prompt length, concurrency or other benchmark conditions.

GreenCore Solutions Corp. CEO Matthew Keddy says gtinfetch checks product identity before an agent orders; the service also offers payment inside the same call.

The October 13th petition says the panel overlooked evidence that the Pentagon punished Anthropic for its public position on AI safety. A rehearing would reopen the appeal over the designation that blocks defense contractors from using Claude on Pentagon work.

Character.AI founders Noam Shazeer and Daniel de Freitas built the chatbot platform after Google. Its October 12th petition lists more than $10.6M owed to its 20 largest unsecured creditors.

A 2-1 panel ruled on September 25th that federal procurement law can cover Anthropic's ability to restrict Claude's functions, even when the company acts without malicious intent. Judge Karen Henderson dissented.

Co-founder Sagar Batchu is extending Speakeasy from API tooling into agent governance. Private access is available to Enterprise customers as Speakeasy enables it organization by organization.

Reactor documents an SDK and API workflow for FastH3-Omni, with five accepted clip lengths from 5.875 to 14.375 seconds at 24 frames per second. MiniMax says the model generates synchronized audio and video from one prompt; the available documentation provides no Omni-specific price or independent quality results.

Taso Du Val is extending Toptal's talent-screening business into model training and evaluation; Toptal says ExpertCrowd generated more than $5 million in recurring revenue before its public launch.

Arena Physica's integration gives Microsoft Discovery agents access to Heaviside for fast field predictions, with conventional solvers available for final checks. CEO Pratap Ranade studied physics and co-founded Kimono Labs, later acquired by Palantir.

The button opens Google's separate agentic development workspace, where agents can work across code, the terminal and a browser. The latest Gemini macOS app is required; 9to5Google reports that Antigravity access through the integration requires AI Ultra.

The model targets coding, enterprise work and cyber defense. Its first rollout is limited to trusted defenders, with wider access planned in phases.

An October 13th X post points to a LiteLLM entry tied to an Azure endpoint, while Microsoft's public documentation still lists Fable 5.1 as the newest Fable model.

Higgsfield's October 13th comparison prices a 300-clip workload at $403.20 through its API. Its maximum individual Ultra allowance is 600 credits short of the modeled need.

Called Personal Agent by Qwen, the invite-based feature is designed to pursue refunds, file flight-delay claims and book services from inside Qwen App.

The Bengaluru AI company introduced its 11-language speech model on September 30th, touting lower latency and more natural delivery for voice applications.

Altair-1 launched on October 1st with NVIDIA processors and Mistral models onboard, beginning a $1 billion program led by Loft Orbital and Marlan Space.

Posted on October 9th, Evo2 pairs code edits with performance feedback to train the model proposing and evaluating those edits.

The multi-year agreement gives the homebuilder access to Promise's factory and robotics service, advancing co-founder Ramtin Attar's push to sell production capacity, not standalone machines.

Matthew Miller's coding-model leaderboard places Muse Wonder eighth of 29 models; its score is an editorial assessment, not a success rate.

Brian Erickson joins after finance roles at DigitalOcean, Amazon and Microsoft as Runpod's company-reported ARR rises from $120M in January to more than $325M.

Co-founders Parth and Kaahan Radia say Persona 2.0 Live is in preview, with emotion tags in prompts and an SDK for finer control.

Morgan Linton says Muse Code limits stopped a full VulcanBench run, raising an early cost-per-task question for Meta's new coding model.

The October 12th test put Meta's new model through two detailed Canvas projects; GMI reported lower costs and faster completion in both runs.